APPLICATION OF A VISION SYSTEM TO THE MONITORING OF CABLE STRUCTURES Elsa CAETANO, Sérgio SILVA and João BATEIRA University of Porto, Faculty of Engineering, Dept. Civil Engineering R. Dr. Roberto Frias, 4200-465 Porto, Portugal
[email protected] Abstract The current paper describes the development and implementation of a vision system in the context of the monitoring of a cable-stayed bridge, discussing the major requirements, the cost of the instruments, the particular aspects of the developed image processing algorithms, the integration in a conventional monitoring system and the results of application to the International Guadiana Bridge in Portugal. This innovative system is shown to constitute a very useful complement to a conventional monitoring system whenever cable measurements are required, allowing not only for the identification of the lower order natural frequencies of these elements, but also for the detection of the start-stop of vibration events and corresponding conditions. INTRODUCTION The current trend of construction of lightweight structures and the immense possibilities resulting from improved material properties and from ever increasing computation power have led to a wide use of cables as structural supporting elements in bridges and special structures. Modern cable supported structures are typically marked by high flexibility, low mass and low damping, characteristics that dictate a significant vulnerability to dynamic excitations, resulting often in cable vibrations. Understanding the different phenomena involved in cable vibrations is the first step in the process of designing and installing control devices to attenuate or prevent these effects. Given the complexity of the problem, intensive research has been conducted worldwide during the last decades, focusing on the identification of vibration events, on the evaluation of patterns of occurrence and on the testing of possible control devices. This investigation is primarily supported by experiments, which can be conducted in laboratory under controlled conditions (comprehending wind tunnel tests), on cable prototypes specially constructed to study particular types of vibration, under partially controlled conditions, or else on structure prototypes. Prototype monitoring of cable vibrations is essential to characterise some phenomena, like parametric excitation. It is also the most challenging type of research, both in terms of instrumentation and implementation of observation programs, and in terms of interpretation. The fact is that cable structures are frequently very large structures, composed by many cables and therefore prototype monitoring means the necessity to employ a large number of sensors and appropriate triggering conditions. Furthermore, since different causes can contribute to the vibration of a single cable, the interpretation of measurements becomes a delicate task and requires long-term observation of the cables under different weather conditions and uses. The difficulties posed in the prototype observation have limited its use and restricted applications. In most circumstances limited measurement campaigns are conducted, and installed monitoring systems usually comprehend no sensor or only a very small number of sensors in cables. New developments in instrumentation and communication systems are however continuously happening that allow re-design of conventional monitoring systems in such a way as to provide the highest amount of data at controlled costs. Considering the particular characteristics of cable
structures and of the monitoring of their dynamic behaviour, it is of interest to investigate the potential of modern vision systems which employ digital cameras as sensors and sophisticated image processing algorithms to transform the motion of selected points within a frame into time series. Compared to conventional monitoring using accelerometers and strain gages, video cameras are lower sensitivity sensors. However, their cost, the easiness of installation, requiring no targets, and the concept of distributed measurements allowed by the spatial coverage of measurements in a particular area are very important advantages that should be considered in conjunction with the knowledge that cable oscillations are normally some orders of magnitude higher than the remaining structural elements, as deck and towers in the case of a bridge. Furthermore, the integration of cameras in a conventional monitoring system provides new features to this system, like the possibility of combining vibration information of all or of a number of cables together with vibration of other structural elements, as the deck and towers, as well as simultaneous meteorological data, namely wind speed and direction. In addition, the on-line image processing of particular points in the observed scene allows the triggering of measurements by means of the video camera whenever a certain level of cable vibration occurs for a number of selected cables. The current paper describes the development and implementation of a vision system in the context of the monitoring of a cable-stayed bridge, discussing the major requirements, the cost of the instruments, the particular aspects of the developed image processing algorithms, the integration in a conventional monitoring system and the results of application to the International Guadiana Bridge in Portugal. This vision system is shown to constitute a very useful complement to a conventional monitoring system whenever cable measurements are required, allowing not only for the identification of the lower order natural frequencies of these elements, but also for the detection of the start-stop of vibration events and corresponding conditions. OPICAL INSTRUMENTION IN CIVIL ENGINEERING APPLICATIONS The use of optical sensors in the observation of civil engineering structures has been very limited and has focused essentially in the measurement of absolute values of displacement. The Portuguese National Laboratory of Civil Engineering, LNEC, has developed one of the first applications of optical instrumentation during the 1960’s, in the context of the construction of the first Tagus suspension bridge.
Figure 1- Automatic device for remote measurement of the displacements [1]
According to the scheme of Figure 1 and to the image in Figure 2, optical systems fitted with servocontrols mounted at the bridge tower pointed to spotlights placed at the ¼ span and midspan of the bridge deck. The motion of these devices along vertical and horizontal direction was converted into an electrical signal that was recorded in magnetic tape or plotted on paper. Marecos et al. [1] stated that the system was stable and not disturbed by fog and other occasional obstructions, and could register minimum displacements of about 0.5cm.
Figure 2- Optical systems in operation at the recording station on the Tagus Bridge [1]
More recent systems have employed digital cameras to record the motion of targets and, using target recognition algorithms, have been directed again to determine absolute displacements of midspan points in bridges [2,3], or else detect the dynamic response at particular sections under traffic load, playing the role of an alternative sensor to a conventional accelerometer [4]. Modern video cameras have equally been used in the characterisation of the motion of dense meshes of points in Laboratory tests [5,6] in situations where contact measurements would not be possible. Videogrammetry techniques are then required in order to accurately characterise the spatial motion of those meshes of points. The concept of distributed sensors, i.e., of single sensors that can capture the spatial deformation, has been introduced in two different ways, one based on laser devices, and the other on high speed cameras. Laser systems are devices that compare a transmitted laser beam with the received beam after reflection by a target or by an oscillating surface. The Doppler-based interferometer vibrometer is a high precision sensor that is particularly adequate for cable vibration measurements and has been successfully used to identify cable forces at the Vasco da Gama Bridge using the vibration method (Figure 3), with a precision comparable to an accelerometer [7]. Considering the operational wide frequency range (DC-300kHz), these devices can be fit with a scanning beam positioning system that enables the spatial characterisation of relative motion, which is not however of interest for cable vibration characterisation, but instead for the characterisation of the motion of the deck and towers, provided that measurement distances are lower than 100-200m. Another possibility for spatial characterisation of motion consists in the use of high-speed cameras. A camera sampling at 600 images/s has allowed the spatial characterisation of vibration modes of a bar in the laboratory [8]. The vision system presented in this paper gathers this concept of distributed measurement together with spatial characterisation of motion described by Yoshida et al. [9], considering however particular aspects that clearly are unfavourable to optical systems: the precision of measurements, the requirement for fixity of the camera and the involved frequency of measurements, and taking advantage of other favourable characteristics: the high quality and low cost of modern digital cameras and lenses, the easiness of installation and the possibility to characterise motion of wide areas within the structure.
Acceleration PSD (m^2/s^4)
1.00E+00 Accelerometer
1.00E-02
Laser
1.00E-04 1.00E-06 1.00E-08 1.00E-10 1.00E-12 0
0.5
1
1.5
2
2.5
3
3.5
4
Frequency (Hz)
Figure 3- Dynamic tests on the stay cables at the Vasco da Gama Bridge, conventional accelerometer vs. VPI sensor for measurements at the same location on the cable
VISION SYSTEM FOR CABLE VIBRATION MEASUREMENT Specifications The vision system presented herein was developed in the context of the monitoring of a cable-stayed bridge where frequent events of cable oscillation were observed, part of them possibly related to a phenomenon of parametric excitation. The identification of the conditions that triggered the isolated vibration of particular stay cables meant the necessity of instrumenting a large number of cables, which was an extremely demanding task in terms of cost of instrumentation and labour in mounting electrical cables. In considering the possible development of a vision system for the purpose of detecting and quantifying cable vibrations in the bridge, the corresponding total length of 666m, a centre span of 324m and a tower height above the deck level of 80m were taken as reference. Then it was assumed that one camera would be mounted on one of the towers’ “legs“ at a particular height, in such a way that the 16 cables of the central span downstream anchored to that tower could be observed. Figure 4 represents the observed scene from deck level at the International Guadiana Bridge. The projected area to observe would then be about 90m distant from the camera. It was accepted that only the projection of the cable vibrations into the plan orthogonal to the direction of mounting of the camera would be accessed, so only one camera was used to characterise motion of each plane of cables. The choice of the camera and lenses was made based on the specification that the system would be able to detect vibrations of cables with diameters in the range 0.15m-0.25m, with amplitudes larger than 0.10m, and with vibrating frequencies in the range 0.7Hz-10Hz.
2 4
Figure 4- View of tower and target cable plane. Plane of cables observed by a camera installed on that tower at deck level
Characteristics of the vision system The vision system is formed by a digital camera acting as sensor. Images are recorded and processed by a laptop, which can also act in conjunction with the camera to program and trigger events. In order to reach a high resolution in the characterisation of the motion, the amplitude of vibration should be characterised by a number of pixels. For the current application, a high quality camera with a resolution of 800x600 pixels and a rate of 30frames/s was selected. This camera is mounted with low distortion lenses and is capable of progressive scanning instead of the typical interlaced scanning common in commercial low-grade devices. This is an aspect of utmost importance, as high frequency components are present in the recorded images and the interlacing introduces artefacts and distortion for these frequencies. The camera has a trigger feature that enables synchronised image acquisition. This synchronisation can be obtained from a master clock, such as a GPS with configurable pulse timing output or some other kind of clock. This enables the correlation between images acquired with different cameras and data from other sources. As the system is designed to operate in outdoor environments, special care must be taken with the behaviour of the cameras with changing lightning conditions. In opposition to common industrial/scientific machine vision applications where the light is set up as to uniformly illuminate in a constant manner the objects of interest, in structural vibration analysis system the light varies greatly within the scene of interest and with time (e.g. the lightning conditions at 9:00 am are certainly different from the lightning conditions at 1:00 pm). Because of this, the camera has to be able to adjust its sensor gain and exposition time to accommodate these variations and provide a steady exposure. The lenses used are standard among machine vision applications; they should have good controlled and easily correctable distortion (e.g. the spatial distortion characteristic must be well approximated by a low order polynomial) and suitable resolution. The acquired data is stored in a hard disk adequate for handling high quality video. The processing unit should support the two operation modes of the system: real time image processing and image recording. The latter is simple to achieve using a standard modern portable computer; the former is however more demanding as the algorithms are of high complexity. Nevertheless real time computation is still in reach of modern portable low power computers. Some care has to be taken in the housing of the cameras/lenses. The housing should be ventilated and the glass window should have a special heating system to prevent condensed water to accumulate in it. Ideally the glass window should have a UV filtering capability and a water and dust repellent coating.
Image processing algorithms and software The implementation of the vision system as a component of the permanent monitoring system, with the possibility of triggering the acquisition of data upon detection of cable vibration, determined the choice of image processing algorithms, which were requested to operate in real time for a number of selected points. Considering the characteristics of vibration, two other requests were added, robustness and high precision. The demands were that the vision system could operate during the day, under varying lightning conditions and in real time, for a sampling rate of 30 frames/s, and could detect very slight motions, defined by only a few pixels in the frame. Considering that, compared to the image area, the relative amplitude of the motion is extremely small, common motion tracking algorithms that rely on some kind of template matching or geometrical property fail. Instead an approximation based on the optical flow algorithm has been selected, that compares two consecutive frames from a video record and defines a vector field describing the intensity changes. If a motion of an object induces changes in the image intensity values then, after suitable camera/lens calibration and after consideration of the point of view, this object movement can be estimated by its optical flow. The optical flow algorithm gives an estimate of the horizontal and vertical optical flow based on the image brightness components. This is made solving the following differential equation:
I x ⋅ u + I y ⋅ v + It = 0
(1)
where I x , I y and I t are the horizontal, vertical and time domain image intensity derivatives, and u , v are the horizontal and vertical optical flow components, respectively. There are many methods to numerically solve this equation. The Lucas-Kanade method gives good results and is suitable for realtime implementation [10,11]. Practical testing has however shown that the Horn-Schunk method [12] enables better results with a low processing speed penalty. The translation of image velocity points, expressed in pixels/s into amplitudes of motion (m/s) can then be made by use of scale factors derived easily from the actual dimensions of structural elements observed with the same angle of the camera. In this way errors of perspective are also corrected. Moreover, the method does not require the previous knowledge of the observed objects, therefore no targets or “training” of the algorithm are required. It is still of interest to refer the possibility to estimate the vibration of the camera from the analysis of static objects present in the image. However, it should be emphasized that it is preferable to avoid those vibrations, as their estimate is noisy.
Figure 5- Graphical interface for the control of acquisition and processing of the vision system
Practical testing of the developed system has shown that computation power available allows the realtime processing of the motion at a maximum of 8 measurement points. The resulting time functions are then visualised and the corresponding power spectrum calculated, which allows the identification
of fundamental frequencies of vibration. The amplitude of oscillation is compared to a trigger level previously established that activates the high precision auxiliary data acquisition system by means of an alarm message. Time series are stored in text files that are classified using a time reference. Figure 5 shows the graphical interface developed for the control of acquisition and processing of the vision system, which includes a low-pass filter to reduce the high-frequency content of the measurements. Integration in conventional monitoring system The interest to correlate cable vibrations with vibrations of the deck and with the velocity and direction of the wind have motivated the purpose of integration of the vision system in the monitoring system designed for the bridge, based on the instrumentation with conventional accelerometers and one anemometer. The designation “integration” is used in this context to express the possibility to obtain synchronised measures of cables and deck/ towers vibrations, as well as interaction between different acquisition systems. The scheme of Figure 6 illustrates the created application. Accordingly, the real time processing of a number of selected points in the image allows detection of the starting of vibration at some cable. That information is sent by internet or cable to the main acquisition system, in such a way this system starts recording data. The measured signals and image are then saved and classified on a database and can be related as they are synchronised via GPS. It should also be noted that a particular wind condition or amplitude of vibration of the deck could also trigger the conventional monitoring system, which in this case activates the vision system.
CCD Camera
Image Analysis System Trigger
Gateway/VPN Database/Control
Internet Gateway/VPN
CAN network
Aquisition System
Gateway/VPN
Figure 6- Scheme of integration of the vision system in bridge monitoring system
Measurement errors and limitations Sampling rate, noise in the image sensor and lightning conditions are among the most significant limitations of the vision system. In general, large civil engineering structures and, in particular, cable structures have natural frequencies well covered by the 0.1-15Hz range of the chosen camera. However, it is important to stress that the measurement frequency limit is in reality much lower, as consequence of the lower sensitivity of the camera to very small amplitudes of oscillation that often characterise the higher frequency vibrations. As for the noise in the image sensor, it is naturally a function of the area observed. Figure 7 represents the variation of the noise level as a function of that area. The ordinate in this figure represents the estimated noise power and exibits a linear trend with area: higher area leads to higher measurement noise. Practical applications have shown that, in the best situations, noise levels in the order of some hundreds of µg/Hz have been obtained, when the noise of a high precision accelerometer is two orders of magnitude lower. The system is therefore only adequate to measure high amplitude vibrations, which are precisely typical of cables. It is still important to mention that adverse weather conditions, like intense rainfall and fog can lead to an interruption of measurements. This is a drawback typical of optical systems and cannot be avoided.
Noise vs. Pixel Scale
7000 6000 )) z H (t qr s g/ µ( e si o N
5000 4000 3000 2000 1000 0
0.1
0.2
0.3
0.4 0.5 Pixel Scale (m/px)
0.6
0.7
Figure 7- Level of noise vs. pixel scale (area of observed image)
Cost of the vision system The vision system here described has an approximate cost of 5000€, which includes a laptop with a data acquisition board, the camera, the lens (wide-angle or zoom), the housing box, tripod and accessories. By comparison, a dynamic acquisition system for observation of 10 cables based on a laptop, acquisition board, a set of 10 accelerometers with medium sensitivity (10V/g) and corresponding conditioning units has an approximate cost of 10000€, added by an additional parcel that is required for the purchase and installation of electrical cables. APPLICATION TO THE INTERNATIONAL GUADIANA BRIDGE The design of the current vision system followed the requirements presented above, which were stated considering the application to the International Guadiana Bridge, located in the Southern part of Portugal close to the boundary with Spain. Aspects related with the practical implementation of the system and to the quality of measures are discussed in this section. Location of the camera The initial idea concerning the location of the camera was to mount it on the upstream “leg” of one tower, at a height close to 20 m above the deck level, and direct it to the plane of cables located downstream anchored to that tower. Figure 4 shows a view of that tower and of the plane of cables of interest, as well as the image observed from deck level with a wide-angle lens, which exhibits a relative large bias and would need a calibration in order to eliminate perspective errors. Figure 8 shows the time records and corresponding power spectra obtained for the 2 points marked in Figure 4, corresponding to the second and fourth longest cables of the central span, with the camera located at deck level. C2
0
20
40
60
80 Time (s)
100
120
140
C2 C4
Amplitude PSD
Velocity
C4
0
1
2
3
4
Frequency (Hz)
Figure 8- Records of motion at the four longest cables of the centre span with camera located at deck level.
5
These measurements were obtained under normal traffic condition, in a situation where the second largest cable (C2) vibrated moderately, and the fourth longest cable (C4) presented a very slight oscillation. The irregularities in the time records represent obstructions introduced by the passage of heavy lorries, which would not occur for a camera located 20m above the deck. It is important to show that, although an accelerometer would detect the various harmonics of the vibration, despite their very low contribution, the vision system hardly evidences the very small contribution of vibration in the higher frequencies than the first mode. The difficulty to access the bridge tower from the exterior in order to mount the camera motivated the selection of a different location, which is a building belonging the National Institute of Conservation of Nature located 850m away from the bridge (Figure 9). From this location, using a zoom lens, the diagonal of the observed area is 300m, instead of the 100m corresponding to the camera location on the tower. Measurements are therefore noisier. However, the bridge perspective is closer to the perpendicular, therefore the distortion introduced in the image is smaller. Portugal
Spain
Figure 9- Aerial photograph of bridge site indicating the location of the camera. View of bridge from that location
Measurements Considering this last location of the camera and using the zoom lens at a focal lenght of 50mm, some experiments were developed covering different areas. Figure 10 shows an image covering an area of 112x150m in which 4 points where selected, corresponding to the third upstream and downstream (3U, 3D), fourth downstream (4D) and fifth downstream (5D) longest cables of the central span anchored to the tower closer to Portugal. That figure shows also the time records obtained after image processing, which are representative of a very common situation where the four longest cables of the central span exhibit significant vibrations, but not equal vibrations occur on the equal upstream and downstream cables. In effect, the vibration amplitudes observed for cables 3D, 3U, 4D and 5D were estimated as 0, 0.14, 0.17 and 0.08m, respectively. Considering the fact that the selected measurement points are not the antinodes of the mode, and assuming the response is dominated by the fundamental mode, the amplitude of vibration of the cable 4D has been estimated as 0.40m. Figure 11 presents the average power spectra associated with the 150s records referred above, partially shown in Figure 10, and Table 1 resumes the peak frequencies of these graphs, which are in agreement with a previous cable measurement campaign using accelerometers. It can also be observed that various harmonics are present in the signals, even for cable 3D, that experiences an extremely low level of oscillation. Moreover, two of the peaks of the average power spectrum density obtained for this cable are associated with global bridge frequencies: the third symmetric global mode, with a
measured frequency of 1.66Hz, and the third anti-symmetric mode, with a natural frequency of 1.88Hz. Vibrações no ponto 1
0.1 ) m ( e d ut i pl m A
0.05 0 -0.05 -0.1
5D 4D 3D
4
6
8
10 Tempo (s) Vibrações no ponto 2
12
14
16
4
6
8
10 Tempo (s) Vibrações no ponto 3
12
14
16
4
6
8
10 Tempo (s) Vibrações no ponto 4
12
14
16
4
6
8
10 Tempo (s)
12
14
16
0.1 ) m ( e
3U
d ut i pl m A
0.05 0 -0.05 -0.1
0.1 ) m ( e d ut i pl m A
0.05 0 -0.05 -0.1
0.1 ) m ( e d ut i pl m A
0.05 0 -0.05 -0.1
Figure 10- Selection of analysed points at Guadiana Bridge. Time records of cable vibration: 3D, 3U, 4D, 5D (from top to bottom) Densidade Médida de Potência Espectral Normalizada (ponto 1) -10 ) B d( N E P D
-20 -30 -40
0
0.5
1
1.5
2 2.5 3 Frequência (Hz) Densidade Médida de Potência Espectral Normalizada (ponto 2)
3.5
4
4.5
5
0
0.5
1
1.5
2 2.5 3 Frequência (Hz) Densidade Médida de Potência Espectral Normalizada (ponto 3)
3.5
4
4.5
5
0
0.5
1
1.5
2 2.5 3 Frequência (Hz) Densidade Médida de Potência Espectral Normalizada (ponto 4)
3.5
4
4.5
5
0
0.5
1
1.5
3.5
4
4.5
5
-10 ) B d( N E P D
-20 -30 -40
-10 ) B d( N E P D
-20 -30 -40
-10 ) B d( N E P D
-20 -30 -40
2
2.5 Frequência (Hz)
3
Figure 11- Average power spectrum density of cable vibration records TABLE 1- Natural frequencies evidenced by the power spectra estimates of cable vibration records Cable 3 Downstream 3 Upstream 4 Downstream 5 Downstream 4 Downstream, accelerometer
1st 0.90 0.92 0.92 0.92 0.90
Natural frequency (Hz) 2nd 3rd 4th 1.70 1.85 2.77 1.85 2.77 3.69 1.85 2.77 3.69 1.85 2.77 1.81 2.71 3.62
5th 3.69 4.6 4.6 4.52
Validation of measurements In order to validate the results obtained with the vision system, additional measurements were conducted on the bridge by instrumentation of both the fourth longest cables of the central span downstream, above designated as cable 4D, and the deck, close to the cable anchorage. These measurements were developed about one hour later, when apparently the level of oscillation of cable
4D was slightly lower, and the other cables exhibited a very slight vibration, the wind blowing from South with a very small angle to the perpendicular to the bridge axis. Figure 12 shows the average power spectrum density estimates (PSD) obtained both with the accelerometer and the vision system time records. It can be observed that the natural frequencies obtained with the vision system are slightly higher than the corresponding frequencies obtained with the accelerometer. This fact is consistent with the usual observation of increased frequencies for higher amplitudes of vibration. It is also observed that the global bridge frequencies captured by the accelerometer in the range 1-1.5Hz are not evident in the video record, as consequence of their very small contribution to the vibration. 1.00E-05
1.E-01 Accelerometer
1.00E-06
Video
1.E-02
PSD Acceleration
1.00E-07 1.E-03
1.00E-08 1.00E-09
1.E-04
1.00E-10
1.E-05
1.00E-11 1.E-06
1.00E-12 1.00E-13
1.E-07 0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
Frequency (Hz)
Figure 12- Instrumentation of cable 4D with accelerometer. Average PSD of accelerations measured using the accelerometer and the vision system.
Monitoring of cable vibrations The use of the vision system with an installed camera as far as 850m from the bridge naturally compromises the detection of very small oscillation of the deck, but it captures the dominant components of oscillation of the cables whenever they reach 8-10cm. This feature is of high interest, considering the low cost of the instruments and small amount of labour in installation and processing data. This system constitutes an excellent tool for remote detection of cable oscillations and will soon be installed at the referred building close to the bridge on a permanent basis, communicating with the University by means of a UMTS broadband connection, transferring short records of data and sending alerts whenever particular levels of vibration are detected at some cables considered critical. This will allow to quantify the frequency and duration of vibrations, as well as to correlate those vibrations with wind speed and direction. CONCLUSIONS The paper focuses on the use of optical devices in the monitoring of cable vibrations. Among the various existing devices, one particular system is chosen that is based on high quality digital cameras which act as sensors and, combined with a laptop, that is required for image processing and for programming and communicating with other conventional monitoring systems, provides these systems a complementary mean to characterise relatively large vibrations of a significant number of cables. Although the precision of this complementary system is around two orders of magnitude lower than that of a conventional system based on high sensitivity accelerometers, the possibility of combining time series resulting from image processing with simultaneously obtained time records based on conventional instruments, opens new possibilities to a deeper characterisation of cable structures. Moreover, for structures where contact measurements are impossible or very difficult, like slender chimneys, antenna towers and power lines, a vision system can constitute an excellent alternative monitoring system.
ACKNOWLEDGEMENTS The present work has been developed in the context of the Project POCTI/ECM/46475/2002: “Vibrations in cable-stayed bridges”, funded by the National Science Foundation FCT. The authors would like to acknowledge this support and also the National Road Institution EP and the Institution for the Conservation of Nature ICNB, for the authorisation to perform the tests and use their facilities. REFERENCES [1] J. Marecos, M. Castanheta and J. Teixeira Trigo, 1969, “Field Observation of Tagus River Suspension Bridge”, Journal of the Structural Division ASCE, vol. 95, no. ST4, 555-583. [2] J. Macdonald, J., E. Dagless, B. Thomas and C. Taylor, 1997, “Dynamic Measurements of the Second Severn Crossing”, Proc. Instn. Civil Engineers, vol. 123, 241-248. Paper no. 11483. [3] A. Wahbeh, J. Caffrey and S. Masri, 2003, “A vision-based approach for the direct measurement of displacements in vibrating systems”, Smart Mater. Struct., vol. 12, 2003, 785-794. [4] Lee, J., Shinozuka, M., 2006, “A Vision-based System for Remote Sensing of Bridge Displacement”, NDT & E International, Vol. 39, 425-431. [5] C. Chang and Y. Ji, 2005, “Flexible Videogrammetric Technique for Three-Dimensional Structural Vibration Measurement”, Journal of Engineering Mechanics ASCE, vol. 133, no. 6, 656-664. [6] G. Fu and A. Moosa, 2002, “An Optical Approach to Structural Displacement Measurement and its Application”, Journal of Engineering Mechanics ASCE, vol. 128, no. 5, 511-520. [7] A. Cunha, E. Caetano and R. Delgado, 2001, “Dynamic Tests on a Large Cable-Stayed Bridge: An Efficient Approach”, Journal of Bridge Engineering ASCE, vol.6, no.1, 54-62. [8] S. Patsias and W. Staszewski, 2002, “Damage Detection using Optical Measurements and Wavelets”, Structural Health Monitoring, vol. 1 (1), 7-22. [9] J. Yoshida, M. Abe, Y. Fujino and K. Higashiuwatoko, 2002, “Image analysis of human induced lateral vibration of a pedestrian bridge”, Proc. of the International Conference Footbridge 2002, Paris, November. [10] B. Lucas, B., T. Kanade, 1981, “An Iterative Image Registration Technique with an Application to Stereo Vision”, Proc. of Imaging Understanding Workshop, 121-130. [11] M. V. Correia, A. C. Campilho, 2004, “A Pipelined Real-Time Optical Flow Algorithm”, Proc. of ICIAR 2004, 372-380. [12] B. Horn, B. Shunk, 1981, “Determining Optical Flow”, Artificial Intelligence, vol. 17, 185-203.
